Fast neighbor user searching for neighborhood-based collaborative filtering with hybrid user similarity measures

In neighborhood-based collaborative filtering (NBCF) algorithms, user similarity measures have a great effect on the performance of collaborative filtering (CF). Researchers have proposed some schemes of hybrid user similarity and applied them to recommendation systems (RSs). However, hybrid user similarity measures suffer from a time-consuming issue when searching neighbor users using these schemes. To solve the issue, this paper proposes a fast neighbor user searching (FNUS) method for NBCF with hybrid user similarity schemes. FNUS first generates three item subspaces: interested item, neither interested nor uninterested (NINU) item, and uninterested item subspaces. In these subspaces, we calculate co-rated item numbers between a target user and other users and then use these numbers to find three subsets of neighbor users for the target user. The final neighbor user set is obtained by finding the union of the three neighbor user subsets. As the calculation of co-rated item numbers is much simpler than that of hybrid user similarity, FNUS can fast search neighbor users. Experimental results on three public datasets show that the proposed method can greatly improve the performance of RSs.